Repository navigation
Coalescing UX: the truth-telling around the unique-open-pair law #45
Description
Activity
Posted by an AI agent on Jackson's behalf
Part 3 resolved by refusal, not by a new surface (Jackson's 2026-09-02 ruling on #44, option D: an agent-to-person send is an ask or a reply; anything else is refused). A second ask from an agent to a person while that pair already has an open exchange is now refused before the ledger write (
A2AHumanFollowupNotAllowedError, PR #57) instead of coalescing as a follow-up delivery — which, for a person, was exactly the invisible-follow-up problem this part describes. The item the person sees is therefore always the whole ask; amending it is loud:clear_own_ask, then re-ask with the combined content. Agent↔agent coalescing is untouched.Record:
docs/j5/product/a2a/agent-tools.md(send_message revision corollary),docs/j5/product/features/inbox.md(sender-side mirror),docs/j5/worklog/human-addressed-sends-ruling-2026-09-02.md— commit 37ef461 on j5/main.Parts 1 and 2 remain open (sender-card 'Awaiting reply' overclaim on coalesced asks; whether the sender is told its ask coalesced) — both agent↔agent concerns.
- added a commit that references this issue
on Sep 2, 2026
Posted by an AI agent on Jackson's behalf
From Jackson's integrated test session (2026-09-01). The unique-open-pair law (one open exchange per sender→receiver pair; later asks coalesce onto it as follow-ups) WORKED correctly mechanically — but every surface around it told an incomplete story:
This is a product design pass over one mechanism's three surfaces — the mechanics themselves are settled law and stay unchanged. Clearing behavior is also settled: one clear/answer closes the shared exchange (verified live).